Attorney-trained AI for commercial insurance. It doesn't just tell you what a policy says - it tells you what the policy means, and shows the citations.
Above: the Qumis mark - a Q drawn as a magnifying glass over a target. Reading the fine print, literally.
Somewhere in a broker's inbox right now sits a 180-page commercial policy tower, three endorsements deep, with an exclusion buried on page 114 that quietly undoes a promise made on page 12. A claim will land on it eventually. And the person who has to explain the gap will do what the industry has always done: read. Slowly. By hand. Praying they catch it.
Qumis was built for that exact moment - the one before the mistake. The Chicago company makes AI that reads commercial property and casualty policies the way a seasoned coverage attorney would: not skimming for keywords, but reasoning about what the language actually does. It compares quotes, binders and endorsements. It flags where a tower leaves a hole. And crucially, it never asks you to take its word for it - every answer arrives with source-linked citations and a visible chain of reasoning.
That last part is the whole company, really. Insurance is a business of defensible decisions. An answer you can't trace is worse than no answer at all. So while much of the AI world raced to build confident chatbots, Qumis built something quieter and harder: a system that shows its work.
Note what it doesn't say. It doesn't say "AI-powered." Half the industry says that. Qumis says "attorney-trained," and the distinction is deliberate. The model didn't learn coverage from the open internet. It learned from thousands of real-world coverage analyses - the accumulated judgment of people whose job was to be right about the fine print when money was on the line.
The origin story is almost suspiciously neat. Dan Schuleman started his career as an insurance coverage attorney - the person who reads the towers by hand and lives with the consequences. He knew the problem in his bones. What he lacked was the machine to solve it.
That arrived in Shiv Sinha, who previously led application development at Goldman Sachs, including work on the Marcus platform - a crash course in building and scaling fintech that regulated institutions will actually trust. Together they are the two halves this problem needed: domain scar tissue and engineering discipline.
Former insurance coverage attorney who experienced the inefficiency of manual policy analysis firsthand - then built the tool he wished he'd had.
Former head of application development at Goldman Sachs, with deep experience building and scaling fintech platforms including Marcus.
Qumis isn't a single magic button - it's a toolkit organized around the things insurance professionals do all day. Each one collapses hours of careful reading into minutes of checked, cited output.
Spot changes across policies, quotes, binders and endorsements - and understand their implications.
Distill what a policy covers and excludes into a clear, structured summary.
Make quick, defensible coverage decisions backed by expert-quality reasoning.
Get answers pulled straight from the policy documents, in plain language, with citations.
Convert policies into completed spreadsheets automatically - no manual re-keying.
Run rapid compliance checks against requirements before anything ships.
SOC 2-grade secure storage with team-wide access and institutional knowledge that compounds over time.
Qumis raised a $2.2M pre-seed in January 2025, led by Armory Square Ventures with MTech Capital, Grand Ventures, Alumni Ventures and BrokerTech Ventures - plus strategic angels including Kin Insurance's Sean Harper. Thirteen months later, in February 2026, it closed an oversubscribed $4.3M seed led by MTech Capital, with American Family Ventures joining as a strategic backer. Total to date: $6.75M.
Bars scaled to round size. Total raised across both rounds: $6.75M.
Enterprise software usually arrives by mandate. Qumis arrived by word of mouth. At NFP - an Aon company - it started with a small team and grew, organically, to hundreds of users. That's not a procurement win. That's people telling their colleagues it works.
Today Qumis counts five of the 15 largest U.S. insurance brokers among its customers, alongside leading specialty carriers and insurance-focused law firms - the most skeptical readers in the building, which is exactly the point.
Closes oversubscribed $4.3M seed led by MTech Capital with American Family Ventures - total funding reaches $6.75M.
Ships a major platform update: refreshed UX, AI-enhanced workflows, and expanded legal-grade analysis covering related contracts and legal references.
Raises $2.2M pre-seed led by Armory Square Ventures to automate policy reviews and claims analysis.
Qumis founded in Chicago by Dan Schuleman and Shiv Sinha.
Plenty of tools will read a document and hand you an answer. Very few will hand you the answer and the paragraph it came from and a confidence signal telling you how sure it is. In insurance, where a wrong coverage call is a lawsuit, that transparency isn't a nice-to-have - it's the reason anyone lets the software near a real file.
Add the security posture - SOC 2 certified, client data private by default and never used to train the model - and you have the rare AI product built for buyers who are paid to say no. Qumis also gets more useful the longer you use it, quietly building institutional knowledge inside each customer's private environment rather than pooling it.
Return to the broker and the 180-page tower. The exclusion is still on page 114 - Qumis didn't rewrite the policy, and it can't. What changed is the reading. The gap surfaces in minutes instead of hiding until a claim. The citation sits right beside it, so the broker can pick up the phone with a defensible position instead of a nervous hunch. The afternoon of squinting becomes a decision.
That's the modest, unglamorous thing Qumis is really selling: not the elimination of the fine print, but the end of being ambushed by it. Insurance is the oldest data business on earth, still run on PDFs and instinct. Qumis is the quiet argument that some of the most useful AI won't announce itself with fireworks - it'll just make sure nobody has to read page 114 alone again.
Watch & listen: Dan Schuleman on the On Point podcast (search "On Point podcast Qumis" on LinkedIn). Product walkthroughs and demos available by request at qumis.com.